What makes the Spotify playlist Borderless borderless? An introduction to the subject of this portolio

The Borderless playlist does not resemble the happiness level from other genres

---

Here, we look at the happiness (valence + energy) in combination with loudness and mode. Regarding these graphs, the Borderless playlist is widely distributed over valence and has energy levels of 0.25 and up. It seems that the higher the energy and valence for Borderless, the louder the music. There are sligthly more songs with major modalities than with minor ones. When comparing the Borderless playlist with the other genres, we observe that energy levels are similar to Pop and Indie, valence levels are similar to Rock, loudness levels are similar to Pop and Rock, and mode is similar to Indie and Pop. The overall happiness distribution resembles Indie and Pop the most and differs the most from Dance/Electronic.

Since the happiness distribution of Borderless is most similar to Indie and Pop, we will focus a bit more on the characteristics of these genres. In the late 1970s, a combination of these genres was born called Indie Pop. "Indie pop is a music genre and subculture that combines guitar pop with DIY ethics in contrast to the style and tone of mainstream pop music. This genre would reflect the underground’s softer, sweeter side, with a greater emphasis on harmonies, arrangements, and songcraft. Listening to the Borderless playlist, we can indeed find these characteristics.

On the other hand, we have Dance/Electronic as genre that differs the most from Borderless. A distinctive characteristic of Dance/Electronic music is the rarity of vocals in the music pieces. Regarding the fact that Borderless is similar tot Indie Pop and there is an emphasis on songcraft and harmonies in Indie Pop, we could predict that Borderless differs from Dance/Electronic. Besides that, Dance/Electronic music pieces usually have a relatively high tempo. More about tempo can be found in the following tab.

Features of Borderless songs in comparison to the other genres: violins, dots, and boxes

---

Here, we look at violin plots showing the distribution of different features for the different genres. The wider the violin at a specific value, the relatively more songs in that genre have that value. The boxes in the violins represent boxplots including the median and quartiles values, and the dots represent outliers.

The median tempo for the Borderless playlist lays around 115 BPM. This median is mostly different from the Dance playlist. Usually, Dance/Electronic music pieces have a relatively high and steady tempo and are around 129-150 beats per minute (BPM). This is observable in the violin plot for tempo. Outside these BPMs, an Dance/Electronic song is prone to be an outlier. Remarkable is that the BPMs for the genres besides Dance/Electronic are very wide-spread with extremely high values (which are not outliers) included. An explanation for this could be that the Spotify API is only able to detect constant BPMs which is applicable to Dance/Electronic music but not to the other genres. Spotify might detect all side-beats and sounds as main beats, resulting in these high smeared values. Therefore, it would be better to especially focus on the boxplots in the violin plot of tempo. We observe that median is closest to Indie and quartiles are closest to Pop. Again, Indie and Pop are most similar to Borderless here and Dance/Electronic most different.

In the previous tab, we observed that energy levels for Borderless were similar to Pop, and Indie. Here, we can observe that the Boderless boxplot belonging to energy is most similar to the ones of Pop and Indie as well. However, when observing the shape of the Borderless violin, it resembles the shape of the Rock violin the most. It appears that the Rock violin is shifted higher, indicating higher energy levels overall than Borderless. A characteristic for Rock music is indeed a high energy behind everything. Borderless does not share this high energy and is overall lower in energy than Rock.

For loudness, the median of Borderless lays closest to the median of Pop, while the quartiles look most similar to the Indie playlist. Also the shape of the violin of Borderless is most similar to the Indie violin. However, it appears that Indie includes a few more silent songs while Bordeless does not have this. Both the violin shape and the boxplot from Borderless are very different from Dance/Electronic. It can be observed that the upper quartile of Borderless has almost the same value as the lower quartile of Dance Party. This could be accredited to the fact that Dance music is generally louder than other music genres to hit hard on the dance floor.

Danceability for Borderless is most similar to Pop and most different from Rock. Remarkable is that the danceability of Hip-Hop is higher than the one for Dance/Electronic, while the name of Dance/Electronic and its definition indicate that it is a genre especially created to dance on.

Speechiness is low for every genre except Hip-Hop which is logical since Hip-Hop music is related to rapping most of the time. It is surprising that the distribution for Speechiness is most similar to Dance/Electronic, while we recently discussed that there are rarely vocals in Dance/Electronic music while there are mostly vocals present in Borderless. However, when listening to a few songs from the Dance/Electronic playlist, we are able to hear vocals. It could be the case that this Dance/Electronic playlist does not share this distinctive characteristic for dance music, indicating that this playlist does maybe not represent the whole genre.

The distribution for acousticness is almost the same as for Rock. It is completely different from Dance/Electronic. An explanation for this is that Dance/Electronic music employs electronic musical instruments, digital instruments, or circuitry-based music technology in its creation, while Borderless songs employ mostly acoustic instruments.

The distribution for liveness is mostly similar to Pop and Indie but is more smeared out for higher values for Borderless.

Valence would be most similar to Rock. However, there seem to be more lower valences in the Borderless playlist than in the Rock playlist, indicating that are a few more negative songs in the Boerderless playlist. What should be noted is that we observed many features of Borderless being similar to Indie and Pop. However, for valence, we can see that values are generally higher. This could indicate that Borderless songs are quite similar to Indie and Pop songs but have a higher positivity overall.

How does the fastest Borderless song and outlier “Abandoned” look like in a tempogram?

---

These tempograms show the tempo analysis from the Borderless song “Abandoned”. Both tempograms are Fourier-based and therefore strongly tend to pick up on tempo harmonics. Sometimes it can be more informative to wrap this into a cyclic tempogram. The right tempogram shows such a cyclic tempogram.

Regarding the violin plot for Borderless from the previous tab, songs with a tempo above 172 are considered outliers.The greatest outlier is the song with a tempo value of 185.975 and belongs to the song “Abandoned” from artist Kit Sebastian. Therefore, these tempograms will probably look different from the rest of the Borderless songs.

Listening to “Abandoned”, it does not necessarily sound that there are a lot beats per minute. Also when looking at the cyclic tempogram (right), it looks like BPM is constantly around 95. For the other tempogram (left), it looks like there are two main tempos which play throughout the whole song: 200 and 400. Since both the ride symbal and the toms are constantly playing at the same time during this song, it could be the case that Spotify recognizes both instruments as beats.

The key distribution of the Borderless playlist is most similar to the Indie playlist and is the only one which lacks the third key completely

---

This histogram shows the distribution of keys across the different playlists.

Comparing Borderless with the other genres, it seems that Indie shows the most similar distribution of keys. However, some keys (1, 5, 7, 11) appear more in Borderless than in Indie. It is remarkable that key 2 is the most apparent one in RapCaviar, while this is not the case in all other playlists. It appears that the keys are generally well-distributed in the Borderless playlist. Only keys 6, 8, 10 have apparent smaller numbers but the rest seems quite evenly divided. A striking founding is that key 3 is not present in the Borderless playlist at all. This key is also rare in the other genres but is present at least. This could be one of the reasons why this playlist is called Borderless.

Chroma of two beyond language Borderless songs compared

---

This visualisation of the Borderless songs Lâcher prise - Kinkhead and Sweets Room - カメレオン・ライム・ウーピーパイ shows the to what extent the chroma features align with each other. Lâcher prise is a French song and Sweets Room a Japanese song. We are not able to display the Japanese characters from the artist in Rstudio but it is possible to click on the link attached to Sweets Room in order to see the artist.

This chroma similarity matrix shows that there cannot be a pattern observed. Both songs are very different from each other in all features. No covers could be found in the corpus used for this project. Therefore, probably all chroma similarity matrices will show no patterns.

The feel&sound from the recent Borderless song ‘Rocky - Still Woozy’, displayed by chroma and timbre features

---

The two self-similarity matrices are summarised at the bar level but have axes in seconds. They illustrate pitch- and timbre-based self-similarity within singer-songwriter Still Woozy’s recently released ‘Rocky’ (19 February, 2021). Studying both matrices helps us to understand the structure of the song. It is remarkable that both matrices are very similar. It seems that only the first seconds of the song have a clearly different timbre than chroma. Furthermore, it seems like there can be no clear structure observed from the matrices while in the song itself, you can clearly hear the the difference between the verses and choruses. The song does not have a fade-out which is observable from these matrices since te pattern at the end is not different from the rest of the matrices.

Classification of the genres by Spotify using Random Forest shows improvement

---

Here it is investigated if Spotify is able to classify the songs in the corpus to the genre to which they belong. This could be hard for songs in the Borderless playlist, since it is possible that they do not belong to any genre.

We starting looking for the most important features Spotify uses for this classification task and found that a variety of timbre features plus duration, danceability, and tempo ranked on top of the list. Using these features, a prediction was calculated which is shown in the heatmap. Then the complete corpus was divided into five parts, using the Random Forest algorithm for cross-validation and to train Spotify to perform better in each run. The calculated precision and recall can be found in the table below the heatmap.

We were able to observe before that Borderless differed most from Dance/Electronic and was most similar to Pop and Indie. From the table should be noted that the precision and recall is highest Dance/Electronic and approximately the same for Borderless, Indie, and Pop. This supports our earlier findings.

The scatterplot shows two timbre components that played an important role in this task.

Borderless shares most similarities with Indie and Pop and least with Dance/Electronic

This project investigated in what aspects the playlist Borderless resembled and differed from the most popular genres Hip-Hop, Indie, Pop, Dance/Electronic, and Rock. We looked at the different features separately and combined in happiness, compared keys between playlists, and tried to improve genre classification with the k-nearest neighbours method and the random forest algorithm. Furthermore, we looked inside the Borderless playlist by comparing two songs from different languages in a chroma similarity matrix and a recently appeared Borderless song with itself.

Comparing the different features (Danceability, Energy, Loudness, Speechiness, Livelenss, Acousticness, Valence, Tempo) separately and combined in happiness resulted in most similarities with Indie and Pop music, and least with Dance/Electronic music. However, for speechiness, the distribution for the Borderless playlist mostly resembled Dance/Electronic which was surprising but can be explained by the incapability to cover the whole genre.

It appeared that Borderless was the only playlist which did not have the third key included. Furthermore, we found that by applying the Random Forest algorithm, classification of genres improved, but the best precision and recall resulted from the Dance/Electronic playlist while the three worst ones resulted from Borderless, Pop, and Indie. This could also be accredited to the similarities between Pop, Indie, and Borderless and the biggest difference with Dance/Electronic.

Overall, we conclude that the Borderless playlist shares most similarities with Pop and Indie which can be merged to the genre Indie Pop. The Borderless playlist is most different from Dance/Electronic. Future research could investigate if the Borderless playlist is actually formed from Indie Pop songs.